Field of the Invention
The present invention relates to a system for estimating the complex gain of a transmission channel.
It can be applied especially to the field of digital radio broadcasting using parallel modulators/demodulators in long-wave, medium-wave and short-wave bands that are amplitude-modulated.
The methods that give the best trade-off between cost and efficiency are based on the use of a parallel modulator/demodulator that is often described as the juxtaposition of a large number N of several hundreds of elementary modulators/demodulators with a low bit rate of a some tens of bits per second, each being locked into its own center frequency. These center frequencies are as close to each other as possible so that the bit rate of the transmitted information is the maximum in the frequency band allotted to the transmitter.
For example, according to the present standards of radio broadcasting, a parallel modulator/demodulator should be capable of working within a total band of 9 kHz comprising 288 carriers spaced out at intervals of 31,25 Hz, each carrier being modulated independently of its neighbors, the carriers being synchronous with one another.
When the transmission channel is unstable, especially in the short waves, it is essential to be able to follow its variations and therefore to estimate its complex gain in amplitude and phase at any point in time and on every carrier so as to be able to use a method of demodulation known as coherent demodulation. This method of coherent demodulation enables the optimum exploitation of high spectral efficiency multi-state modulations, namely modulations characterized by a large number of bits/s transmitted per Hz of occupied band.
The complex gain of the channel is usually estimated by inserting symbols of known amplitudes and phases according to a predetermined regular pattern. These symbols are also called xe2x80x9cgain referencexe2x80x9d symbols. They represent the smallest possible proportion of the totality of the transmitted symbols in order to maximize the useful bit rate.
The complex gain of the channel at any point in time and on any frequency can then be estimated by a method of interpolation that uses the gains measured on the gain references to compute the gain of the channel at the desired position.
The method is comparable to that of a filtering operation, and consists in computing a weighted sum of the gain references close to the cell considered.
A first known approach consists in carrying out a temporal interpolation followed by a frequency interpolation. Two steps are then necessary.
The first step consists of the estimation, by interpolation, of the complex gain of the channel on a specified symbol of each of the carriers from the gain references located in the past and the future. This interpolation is done by a linear combination of a specified number Kt of gain references.
The second step consists of a filtering operation done along the frequency axis by means of a transversal filter in order to improve the estimation. All the gains estimated at a given point in time on the carriers close to a given carrier are combined with the gain estimated on this carrier in order to improve the signal-to-noise ratio. This interpolation takes place on Kf carriers.
In the first approach, N is the number of carriers, and the complexity of the set resulting therefrom is proportional to Nxc3x97Kt+Nkf=N(Kt+Kf).
A second approach consists in carrying out a 2D interpolation. In this case, the estimation takes place in only one step, and is the result of a combination of temporal interpolation and frequency interpolation. The gain estimated at a given point in time on a carrier under examination is a linear combination of the gains of the gain references that are present in the past, present and future of the point in time considered, on the carrier examined and the carriers in its neighborhood.
To obtain levels of performance close to those of the first approach, it is necessary to consider slightly less than Ktxc3x97Kf gain references at each interpolation in keeping only those that are xe2x80x9cthe most correlatedxe2x80x9d with the gain to be estimated, which is equivalent to a complexity of about xc2xexc3x97Nxc3x97Ktxc3x97Kf.
A third approach consists in xe2x80x9cmaking a projection on eigen-vectors. In this case, the received signal is deemed to result from the sum of R replicas of the original signal transmitted on a multipath channel. All the gains of the channel considered at a given point in time and on N carriers may be reduced to a gain vector that is the sum of the following vectors: R slowly evolving (narrow-band) vectors that are decorrelated in varying degrees and a random noise vector.
The method used consists in computing the autocorrelation of the gains along the frequency axis, forming a matrix that has these autocorrelations as coefficients, finding its eigen-values and keeping the R eigen-vectors that correspond to the R paths.
The vector of the gains is then projected on these eigen-vectors and the projection is considered to be the totality of the smoothened gains rid of the noise.
The complexity of this method corresponds to the computation of the N autocorrelations, followed by the computation of the eigen-values of a matrix with a size Nxc3x97N. In other words, the complexity is very great.
Finally a fourth approach implements a method of adaptive filtering that uses one of the two methods of interpolation, namely temporal interpolation followed by frequency interpolation or 2D interpolation, and that varies the coefficients of the interpolator filters in the course of time according to well-known mathematical methods such as the gradient algorithm, the simplified gradient algorithm, the sign algorithm, or even the method in which the same computation is repeated at regular intervals. In this case the degree of complexity is between that of the fixed interpolation method and that of the projection method. It entails risks of instability in channels that are-highly disturbed as is the case with the HF short-wave band.
The goal of the invention is to provide an sub-optimal approach that seeks to obtain levels of performance comparable to those of the most cumbersome methods referred to here above. At the same time, it seeks to obtain a method of very low complexity. This is an indispensable condition if the receiver is to have an acceptable cost of acquisition and use, given that the channel is not always disturbed in the same way. The reception conditions of the various broadcasting stations that can be received at a given point in time are indeed highly variable. Some stations come in xe2x80x9cloud and clearxe2x80x9d, while others are affected by noise and distortion. Thus, it would seem obvious that the work to be done by the channel estimator of the receiver will not be the same in both cases.
Worse still, it can be seen that an interpolator adapted to a disturbed channel lowers the quality of the performance on a high quality channel. This can be explained by considering the interpolator to be behaving like a low-pass smoothing filter that must be broad when the channel is unstable and rather narrow when the channel is stable. Should the filter be adapted to an unstable channel while the channel is stable, then this filter will receive excessive noise in addition to the useful signal represented by the gain references received. If, on the contrary, the filter is adapted to a stable channel while the channel undergoes quick change, then it will become impossible to follow its changes so that, if there is little noise, the channel will be badly estimated and the quality of the estimator will be poor.
It is thus advisable, at all times, to have interpolators sufficiently adapted to the current situation which depends on the time and on the station from which reception is being made. In this case, the performance could be of lower quality than in the case of the optimal interpolators but sufficient for the deterioration to go unnoticed.
To this end, another object of the invention is a system for the estimation of the complex gain of a transmission channel designed for parallel type modulators/demodulators that can be used in digital radio broadcasting, the system comprising a given number N of sets of interpolators to estimate the gain on a useful symbol from known symbols constituting gain references evenly positioned in the transmitted signal, each set of interpolators being adapted to a particular situation of reception, ranging from a stable single-path channel to a channel with high temporal dispersion and high instability, the outputs of said interpolators being coupled to a selection device programmed to choose the set of interpolators that gives the weakest estimated noise during a specified length of time.
The invention has the advantage of great simplicity of implementation, the algorithms used for the computation of the selection criteria being similar to those normally necessary to carry out a demodulation. The increase in the computing power necessary is low because the selection criteria are computed only on the gain references. The invention also has the advantage of proposing lower-cost adaptation to extremely variable conditions of reception. Finally it makes it possible to obtain quasi-optimality, namely little deterioration of performance as compared with the performance that can be obtained with a constantly optimal system.